How mansoorkhandurrani/vision-classify-cnn works
Vision Classify CNN trains and evaluates scene classifiers, then serves the MobileNetV2 model through a Flask upload-and-predict web app. The user workflow runs from selecting or dropping an image in the browser, through the prediction endpoint’s image transforms and model inference, to displaying a class and confidence. The repository also includes dataset inspection, augmentation, scratch-CNN training, transfer learning, experiment runs, and confusion-matrix evaluation. Training and evaluation connections follow the shown imports and calls; the README documents the page and sample-image capabilities.
Main entry points: train.py CNN training · app.py Flask prediction API · script.js Upload interface · transfer_mobilenet.pth Saved model weights
Components 14
Web prediction
- Upload interface
app/static/script.js - Web page
app/templates/index.html - Flask prediction API
app/app.py
Training
- Dataset inspection
src/inspect_data.py - Image transforms
src/preprocess.py - Scratch CNN
src/model.py - CNN training
src/train.py - MobileNet fine-tuning
src/train_transfer.py - Training experiments
src/run_experiments.py
Evaluation
- Model evaluation
src/evaluate.py
Model artifacts
- Saved model weights
models/transfer_mobilenet.pth - Evaluation reports
results
Other components
- Image user
- Scene image dataset
Connections 17
- Image user → Upload interface: selects image
app/static/script.js - Upload interface → Flask prediction API: posts image
app/static/script.js - Flask prediction API → Upload interface: returns prediction
app/app.py - Upload interface → Image user: shows result
app/static/script.js - Flask prediction API → Web page: renders page
README.md - Flask prediction API → Saved model weights: loads weights
app/app.py - Dataset inspection → Scene image dataset: inspects images
src/inspect_data.py - CNN training → Scene image dataset: loads images
src/train.py - CNN training → Image transforms: uses transforms
src/train.py - CNN training → Scratch CNN: instantiates
src/train.py - CNN training → Saved model weights: saves weights
src/train.py - MobileNet fine-tuning → CNN training: loads data
src/train_transfer.py - MobileNet fine-tuning → Saved model weights: saves weights
src/train_transfer.py - Training experiments → CNN training: runs training
src/run_experiments.py - Model evaluation → CNN training: loads data
src/evaluate.py - Model evaluation → Saved model weights: loads weights
src/evaluate.py - Model evaluation → Evaluation reports: writes reports
src/evaluate.py